Dr. John Tenney Director of Research Restoration Robotics...

Dr. John TenneyDirector of ResearchRestoration Robotics, Inc.

Robot Control for Medical Applications and Hair Transplantation

Presented to the IEEE Control Systems Society, Santa Clara Valley18 November 2010

Telerobotics Intuitive Surgical – Laparoscopic surgeryHansen Medical – Catheter guidance

Robot AssistMako Surgical – Orthopedic surgery

Fully AutomatedAccuray – RadioSurgeryRestoration Robotics – Hair Transplantation

Company: Intuitive Surgical, Sunnyvale, CA

Product: Da Vinci Teleoperated Multiple Robotic arms Various instruments

3x3 rotation matrix from camera to tool tip is identical to 3x3 rotation matrix from eye to handle.

Supervisor(Finite State Machine)

logical, on demand only(event driven)

Middleman

numerical, on demand only

Kernel

numerical, continual

BPs

Peek/Poke

FRL

ErroreventsNormal

events

Button and otherinterface events

Visual displayand sounds

Numerical data(sensors, config files)

SensorsActuators

incl.MiddlemanResponseevents

•Finite state machine controls UI and app logic for reliable operation.

•Low-level high-performance distributed control systems.

Company: Hansen Medical, Mountain View, CA

Product: Sensei Catheter System Teleoperated distal

control of catheter

Up to 2.8MM CatheterOuter GuideInner Guide

max 15 cmRALA

3.8 MM OD2.8 MM ID

4.7MM OD4MM ID

Company: Mako Surgical, Ft. Lauderdale, FL Robot-assisted jigless partial-knee surgery

12

Fast recovery

Smaller incisions

Dynamic planning changes

Surgeon assist

Company: Accuray, Sunnyvale, CA

Product: CyberKnife Noninvasive, fully

automated.

Many beams crossfire Frame based Limited to cranium Isocentric treatments Painful

Company:Restoraton Robotics, Mountain View Fully automated hair dissection

Current techniques are capable of excellent aesthetic, natural results

Disadvantages Labor intensive, tedious, long Staff recruitment, training, retention No standardization Invasive (strip excision) Stigma of past techniques, results Limited knowledge of procedure by

potential patients Credibility of hair restoration field Source: Bosley.com

More refined approach

Hair follicles naturally grow in clusters (F1, F2, F3+) - follicular units (FUs)

FU grafts are small (~1mm dia.)

Enable an aesthetic approach F1, F2, F3 distribution Angles, patterns, spacing of FU

implants

Follicular unit grafts

Robotic Arm

Mechanism

Cameras

NeedlesForce

Sensor

User Console

Standard Robot Control: Staubli TX60 robot with off-the-shelf controller.

Industrial PC with Windows 7 for high-level machine control, with embedded controllers in robot, power distribution, and needle mech.

Visual Servo Control: Two pairs of stereo cameras—one 35mm FOV, one 15mm FOV—running at 50Hz. Considering move to 150Hz.

Force Control: Six-axis force sensing used for gross positioning by physician.

DesiredPosition(Camera)

Error(Camera)

Transform

Error(Robot)

P-Dcalcs

VelocityCmd

Robot Controller

Camera(25ms delay)

Measured Pos(Camera)

Goal: To provide 100 micron dissection accuracy at 16 hairs/min.

Disturbances: Patient breathing and other motion Delays due to

• Image acquisition and transfer to PC• Image processing• Communication with robot• Robot trajectory generation (25Hz low-pass trajectory filter)

Image Capture

Recti-fication

IEEE1394B

Image Enhancement

• Rectification simplifies the ‘correspondence problem’ by undistorting and transforming images into a common plane

• Image enhancement is performed to minimize effects of glare and shadow.

• Contour analysis establishes outlines for 3-D hair creation and measurement.

• Optical flow is used to assist in tracking hairs across frames.

• Using OpenCV for image processing.• No longer using CUDA for GPU-accelerated vision

processing.

Contour Analysis

Optical Flow Analysis

3-D Modeling and Tracking

Micro FOV

Wide FOV

• Fiducial markings on skin tensioner used to plan back-and-forth path across tensioner.

• Harvesting occurs at a rate of about 15 hairs/minute.

Confidential

3-D Model of robot, needle mechanism, and cart used in real-time collision prediction and avoidance.

Software protection to avoid measured location of head.

Several mechanical touch sensors on needle mechanism.

Force sensor on needle mechanism. Limited depth of field of vision system. Standard E-stop and EPO switches.

Continuous feedback Cameras measure the distance between the needle

and the scalp at every frame Cameras have limited depth-of-field, robot cannot

automatically move past scalp plane. Redundant safety devices

Force Sensor E-stop buttons User overrides

Watchdog checks throughout software and electronics

Clinical Trials Successfully

concluded FDA Clearance

510k submission complete

Product Release in 2011

Medical Robotics is an exciting and growing field. Mechanical Design, Control Systems, and Software Engineering are crucial in making these devices successful.

Restoration Robotics is making a key contribution to the world of medical robotics: first product to fully automate surgery with patient contact

Stereo vision, faster CPU’s and faster camera technology makes visual servoing increasingly feasible in a variety of applications.